Who Benefits from Government Guarantees: Evidence from Saitdol Loans^†

A. Background of the Introduction of Government Guaranteed Loans

Medium-interest rate credit loans are generally defined as products sold to borrowers with credit ratings ranging from 4 to 7, with interest rates ranging from approximately 5% to 18%. The background of the medium-interest rate credit loan market becoming a major topic in policy finance since the end of 2015 lies in the fact that despite the presence of financial consumers with intermediate levels of risk, there was a lack of financial products available in the market at appropriate interest rates. As a result, some of these consumers either gave up on obtaining credit or were pushed towards high-interest-rate loans designed for subprime borrowers.

Figure 1 displays the average interest rates by credit rating and financial business sector as of the end of September of 2015. According to the figure, prime borrowers with credit ratings of 1 to 3 were mostly able to access relatively low-interest-rate credit products offered by first-tier financial institutions. However, a significant portion of borrowers with credit ratings of 4 to 7 had access only to high-interest-rate loans, typically around 20%, offered by specialized lending companies and savings banks.

As a result of this interest rate disparity, subprime borrowers either had to give up on obtaining loans or were compelled to utilize high-interest-rate loan products, as noted above. Table 1 illustrates the trends in the supply of personal credit loans based on year-end data for 2012 and 2015. During this period, the overall scale of personal credit loans increased, but the loan volume for moderate-to-low credit borrowers, excluding prime borrowers, decreased. This indicates that the decreased loan volume for moderate-to-low credit borrowers was not due to a decrease in demand but rather a result of their limited market participation due to the lack of appropriately priced loan products, as mentioned earlier.

The emergence of the homogenization phenomenon in the mid-interest-rate loan market can be attributed to the lack of active supply incentives from financial institutions, coupled with a severe information asymmetry problem between financial institutions and borrowers.7

Therefore, in order to address the market failure caused by information asymmetry, the Financial Services Commission (FSC) introduced a guarantee-backed loan product called Saitdol loans in July of 2016, led by banks.8

Saitdol loans are mid-interest-rate loans supplied by banks, savings banks, and mutual finance companies, with the principal guaranteed by Seoul Guarantee Insurance.9 Financial institutions, such as banks, pay insurance premiums to Seoul Guarantee Insurance when issuing Saitdol loans, and in the event of a borrower default, Seoul Guarantee Insurance compensates the financial institutions by providing insurance payouts. The insurance premiums, in this case, are applied differentially based on the borrower’s creditworthiness, through negotiations between Seoul Guarantee Insurance and each financial institution.10

The insurance premiums and interest rates for Saitdol loans are differentiated based on each financial sector. As shown in Table 2, the insurance premiums and interest rates in savings banks are set higher compared to those at commercial banks. This reflects the intention to address the disparity in interest rates further by diversifying the sales channels based on factors such as the target borrowers and operating costs specific to each sector.

The Saitdol loan, offered under the Saitdol Ⅰ category, is designed for salaried individuals, self-employed professionals, and pension recipients who meet specific income and tenure criteria. Salaried individuals must have an annual income of at least KRW 20 million and a minimum employment period of six months. Self-employed individuals are required to have an annual income of at least KRW 12 million and a business tenure of at least one year. Pension recipients must have been receiving pensions for at least one month.

The loan application process can be conducted either in person at bank branches or through designated mobile platforms. For in-person applications, 13 participating banks across the country offer this service at their branches. Applicants need to provide the necessary documentation to prove eligibility, and upon a successful credit evaluation, loan disbursement can occur immediately.

For those applying through mobile platforms, the process is available via the online platforms of Shinhan Bank and Woori Bank. Applicants must initially register for internet banking and proceed with the Saitdol loan application. The loan is executed upon the completion of guarantee and credit evaluations conducted through the platform.

In cases where income can be verified using health insurance contributions or national tax payment records, same-day loan executions are possible. However, if additional income verification documents are required, the applicant must visit a bank branch, which may extend the loan processing time by one business day. This dual-channel application process ensures accessibility and efficiency while maintaining stringent credit assessment standards.

Saitdol Ⅱ loans, offered by savings banks, are tailored to specific target groups such as individuals who failed to qualify for bank loans, users of high-interest loans in the second-tier financial market (with interest rates of approximately 20%), and those utilizing microloans of up to KRW 3 million. The product is divided into three types to cater to these groups, providing a customized approach to lending.

Compared to Saitdol Ⅰ, the income requirements for Saitdol Ⅱ are less stringent. Salaried individuals need an annual income of at least KRW 15 million, while self-employed individuals and pension recipients must have an annual income of at least KRW 8 million.

The application process varies by product type. For the product designed for individuals who failed to secure bank loans, applications are primarily processed through channels connecting banks and savings banks. The second-tier financial market complementary product allows applications both online (via internet and mobile platforms) and offline (at branch offices). In contrast, the microloan rapid product is exclusively available through non-face-to-face channels, such as internet and mobile platforms.

Savings banks can choose which product type to offer based on their operational strategies and business environments. For in-person consultations, applicants can visit one of 30 savings bank branches nationwide. After submitting the necessary documents, the process involves a guarantee insurance assessment, internal approval by the savings bank, and the completion of the required documents, such as the loan agreement, before disbursement.

For non-face-to-face consultations, customers can contact the respective savings bank’s call center. Online applications can be submitted through each bank’s website or mobile app. After the loan assessment, identity verification is conducted via the call center, followed by the completion of an electronic loan agreement. Once these steps are finalized, the loan is disbursed. This dual-channel approach provides flexibility and accessibility while ensuring compliance with regulatory and credit standards.

B. Current status of the mid-interest-rate loan market

As shown in Table 3, since the introduction of Saitdol loans in 2016, the supply of mid-interest-rate loans, including private sector loans, has consistently increased. As of 2018, commercial banks and savings banks provided a total of around 1.8 trillion won in Saitdol loans, with interest rates at approximately 7-8% for commercial banks and mutual savings banks and around 17% for savings banks. Private sector mid-interest-rate loans also accounted for around 4.1 trillion won, resulting in a total supply of approximately six trillion won in mid-interest-rate loan products.11

Another notable point from Table 3 is still the significant role of the government in the mid-interest-rate loan market. It is noteworthy that Saitdol loans, which are guaranteed loans, account for approximately 29% of the total supply of middle-interest-rate loans based on cumulative supply. The Financial Services Commission also provides incentives for commercial financial institutions to handle private moderate-interest-rate loans through measures such as the recognition system for regional loan amounts in savings banks, in addition to the guarantee loan system.12

Table 4 presents the supply amounts of mid-interest-rate loans according to the financial sector. Savings banks and specialized credit finance companies primarily drive the supply of mid-interest-rate loans. However, the proportion of mid-interest-rate loans among household loans, excluding mortgage loans, is only about 0.82% as of 2018. Looking at the sectors individually, specialized credit finance companies, banks, and mutual finance companies, excluding savings banks (9.41%), still show relatively low handling rates.13 Considering that the outstanding balance of personal credit loans for mid-to-low credit borrowers was 92.9 trillion won as of September of 2017,14 it is a stretch to consider the supply volume of mid-interest-rate loans as sufficient.

The Saitdol loan program demonstrated a steady supply of approximately 2 trillion KRW annually, with total disbursements of 2.1 trillion KRW in 2019, 2.0 trillion KRW in 2020, and 2.0 trillion KRW in 2021. However, concerns have been raised regarding the allocation of these loans, as nearly 70% of the loans intended for mid-credit borrowers were supplied to high-credit borrowers, diverging from the original policy objectives.

Notably, the proportion of high-credit borrowers receiving Saitdol loans increased significantly over time. By transaction volume, their share rose from 13% in 2018 to 34.7% in 2019, reaching 45.8% as of 2020. Similarly, in terms of loan amounts, the proportion allocated to high-credit borrowers grew from 16.8% in 2018 to 39.6% in 2019 and further to 53.6% in 2020.

These trends are closely related to the central question of this study: whether Saitdol loans have been effectively provided to borrowers who would otherwise be unable to secure financing in the private financial market without government guarantees. The substantial allocation of loans to high-credit borrowers raises concerns that public policy financing may have been directed toward borrowers who were not in critical need of such support.

A. Theoretical Background

A unit mass of financial customers who want to borrow money from a financial institution exists. We assume a two-period time-separable utility function of a customer (u(c₁, c₂) = u(c₁) + βu(c₂)). Accordingly, each customer determined the consumption profile of the first and the second period based on their income stream. Moreover, we suppose that the period-utility function u(c) is twice continuously differentiable (C² function).

Information asymmetry between banks and agents stems from uncertainty in the second period income of each borrower. In the first period, the agent certainly earns w₁. However, in the second period, the agent earns when the good state is realized with probability (1 − p), while she earns when the bad state is realized with probability p.15 The probability p is privately known by customers and is distributed with a cumulative distribution function (c.d.f.) F(p | X) in the population. The distribution, F(p | X), is common knowledge. Here, X is the set of observable information that the banks can use to price loans, such as credit scores and the income of each customer, among other information. If the good state is realized, consumers repay the debt, while the loan cannot be paid back if the bad state is realized.16

We impose no restrictions on F(p) ; it may be a continuous, discrete, or mixed distribution, with have full or partial support, and is denoted by Ψ ⊂ [0, 1]. Throughout the paper, an uppercase P denotes a random variable that represents a random draw from the population (with c.d.f. F(p)); a lowercase p denotes a specific customer’s probability of realizing a bad state (i.e., their realization of P).

Customers have standard von Neumann-Morgenstern preferences with the expected utility given by

Here, the contract presented by a financial institution to each financial consumer consists of a pair made up of loan limit L and loan interest rate R.

c₁(p) indicates the first period’s consumption, while and separately indicate the second period’s consumption when the bad state is realized and the good state is realized.17

Under these circumstances, the set of implementable allocations is defined as follows. The set of implementable allocations is a set of allocations that satisfy the three conditions; 1) resource feasibility, 2) incentive compatibility, and 3) individual rationality. Resource feasibility determines whether the consumption level of each group is equal to or less than the expected income level. On the other hand, incentive compatibility determines whether each borrower covets the terms of the contract presented to a borrower with a different probability of default. Lastly, individual rationality determines whether it is reasonable for each borrower to participate in a loan contract instead of consuming the endowment.18

B. Empirical Methodology

The following information is needed to estimate the minimum pooled price ratio and the lower bound of the magnitude of private information discussed above:

Let D denote a dummy variable for default cases overdue by more than 90 working days.

Let Z denote the forecasted probability, which estimates the true probability P that financial institutions (banks) could not observe. Z is information not reflected when determining loan interest rates and the loan amount limit because financial institutions cannot observe it.

Let X denote a set of information that could be observed by the financial institutions and that is reflected in loan screening and pricing.

In this study, we do not make the constrained assumption that the estimated probability Z is equal to the true probability P. Instead, we make two assumptions that are more realistic.25

Assumption 1. The estimated probability Z does not contain additional information other than the true probability P pertaining to default cases.

Assumption 1 implies that Z, the estimated probability of P for each financial agent’s default due to a bad income shock in the second period, could not contain more information than the true probability P with regard to the estimation of the probability of a bad income state occurring.

Assumption 2. The true probability P is unbiased.

Assumption 2 states that the true probability P is unbiased. Under Assumptions 1 and 2, the estimated probability Z for P can be seen as the ‘garbling’ type of the sort of true probability mentioned in Blackwell (1953).26

In order to calculate the pooled price ratio under the above assumption, the distribution of private information should be estimated beforehand. Thus, we estimate the probability density function f_p(p) of private information. Because the data used in this study contain information about the ex-ante forecasted probability Z in addition to information about default cases D, we define the likelihood ratio function based on the joint probability distribution f_D,Z(D,Z).

Here, f_Z|P(Z | P = p) is the distribution of the ex-ante forecasted default probability Z for borrowers given with default probability p. The first equality follows by transforming the conditional expectation. The second equality follows by expanding the joint density of D and Z given P, and the last equality follows from Assumptions 1 and 2.

After specifying functional forms of f_Z|P(Z | P;θ) and f_p(p;v), the maximum likelihood ratio is estimated from the data on default cases D and the predetermined probability Z. Here, θ and v could be estimated using the maximum likelihood ratio, and functional form for f_p(p), the probability distribution of private information, could also be completed.

Once f_p(p) is estimated, the pooled price ratio can be calculated from Definition 2. Theoretically, T(p) can be obtained at any p, but as p increases under finite dataset E[P | P ≥ p] will rely on a smaller and smaller sample size. Thus, the minimum value of T(p) is not well identified over p that has the largest value of the domain of P. As a solution, we present , which is the minimum of T(p) over the restricted domain. We set τ to 0.95 in this study, but the estimates are generally maintained even if τ is changed.

The process for estimating the lower bound of the magnitude of private information is as follows. Let m_z(p) = E_Z|X[P_Z | P_Z ≥ p, X] - p, which leads to Proposition 1.27

C. Relationship between the Theory and the Empirical Analysis

Corollary 2 and Corollary 3 provide the theoretical foundation for comparing the degree of information asymmetry inherent in two groups. To understand the minimum pooled price ratio proposed in Corollary 2 intuitively, it is essential to consider observationally identical groups.

An observationally identical group refers to a set of individuals for whom all observable characteristics have been fully accounted for, meaning no discernible differences remain based on the available observable data. In other words, these groups appear indistinguishable based on observable factors, isolating any remaining differences to unobservable characteristics, which are often the source of information asymmetry.

At this point, we consider two groups, A and B, composed of observationally identical individuals within each group. While groups A and B are distinct due to observable characteristics, individuals within a single group are indistinguishable based on observable data.

In this context, the minimum pooled price ratio values for each group are denoted as T_A(p) and T_B(p), respectively. Here, T_A(p) represents the minimum cost that a borrower in group A with specific default probability p must bear. This cost accounts not only for their own risk but also for the potential costs arising from the presence of other borrowers within group A who have a higher default probability than themselves. In essence, T_A(p) reflects the borrower’s share of the collective risk within their group.

f the minimum pooled price ratio (T(p)) is higher in group B than in group A, it indicates that borrowers in group B face more severe information asymmetry compared to those in group A. Consequently, the likelihood of a market failure is greater for group B borrowers than for group A borrowers. In other words, a higher minimum pooled price ratio suggests a greater degree of information asymmetry between borrowers and financial institutions, increasing the risk of market inefficiencies.

Therefore, if the minimum pooled price ratio for the group receiving private credit loans without government guarantees is higher than that of the group receiving Saitdol loans, the implication is that the information asymmetry in the Saitdol loan group is less severe. This would suggest that, contrary to the government’s intentions, some Saitdol loans were issued to borrowers who did not actually require government-backed guarantees. This misallocation points to inefficiencies in targeting and resource allocation within the Saitdol loan program.

Similarly, the magnitude of private information presented in Proposition 1 is another measure of the degree of information asymmetry. The interpretation of m(p) is as follows: it represents the difference between the conditional mean default probability of borrowers with a higher default risk than p and the specific default probability p within an observationally identical group. Essentially, m(p) will have larger values if a significant number of borrowers with higher default risks than p are concentrated within the group.

This metric enables the measurement of the degree of information asymmetry inherent in two different groups, A and B, composed of observationally identical individuals. While the method proposed in Corollary 3 can estimate m(p) non-parametrically, it does not allow for a comparison across all values of p. Therefore, as suggested in Proposition 1, the analysis focuses on comparing the expected values of m(p). In contrast, Corollary 2 employs a parametric estimation, allowing direct comparisons between the groups and providing complementary results for robustness checks.

This study utilizes both approaches to compare the levels of information asymmetry between the groups. If the degree of information asymmetry in the Saitdol loan group is found to be lower than that in the group receiving private credit loans, it implies that some government-backed loans were distributed to borrowers who did not require such guarantees. Furthermore, if the degree of information asymmetry between Saitdol loan holders and financial institutions is not severe enough to cause a market failure, this also suggests that government-backed loans may have been provided to borrowers who did not actually require such guarantees. These findings would highlight a misalignment in the allocation of government-guaranteed loans, indicating inefficiencies in targeting the intended beneficiaries.

D. Data

The data used in this paper are from one of the largest credit rating agencies in South Korea, covering personal credit loan data. The data consist of a 20% sample of newly originated personal credit loan accounts in the banking and non-banking sector from January of 2013 to December of 2017, including information on Saitdol loan accounts during the same period.29

The data are divided into several categories, in this case borrower characteristics, loan contract details, and delinquency records. The borrower characteristics include credit ratings, credit scores, occupation, age, income, debt-service ratio (DSR), outstanding loan balances, and whether the loan is a Saitdol loan or not, among other factors.

For the empirical analysis in this paper, borrowers are classified into three main groups. The first group consists of borrowers with only bank credit (private, non-guaranteed) loans, representing high-quality borrowers. This group of borrowers is characterized such that each financial institution can sufficiently handle the risk of a default without the need for government guarantees, relying solely on their creditworthiness. Therefore, this group serves as the benchmark to assess whether the information asymmetry level among Saitdol loan borrowers is severe enough to cause a market failure. The second group includes borrowers with Saitdol loans from the banking sector, and the third group comprises borrowers with Saitdol loans from the non-banking financial sector (savings banks).

In order to ensure a fair comparison, the empirical analysis utilizes data from the period after the introduction of Saitdol loans (starting in July of 2016) for all three groups. Additionally, for robustness checks, an analysis using data from before the introduction of Saitdol loans (from January of 2013) is conducted. Detailed results of this analysis can be found in Chapter 5.

Table 5 presents the descriptive statistics for the three groups after the introduction of Saitdol loans. As expected, the group with only bank credit loans exhibits the highest income level, the best credit scores, and the lowest average delinquency rate.

One noteworthy variable related to Table 5 is the Forecasted Default Probability. The Forecasted Default Probability is calculated by the credit rating company using all information it possesses to estimate the default probability for each borrower. This information is not available to other financial institutions in the market. Unlike other financial institutions, the credit rating company collects information from all financial institutions. Therefore, it is a value that incorporates all individual borrowers’ debt structures, consumption information, and other factors that cannot be reflected in the individual evaluation of each financial institution. This Forecasted Default Probability serves as the predicted probability Z.

To qualify as a predictor Z, two primary conditions must be met. First, Z must incorporate information that is neither directly considered nor accessible during the loan evaluation process conducted by individual financial institutions. Second, Z should represent an estimate that includes only unbiased errors (white noise) associated with the actual default probability (P). Given that the Forecasted Default Probability, as calculated by credit rating agencies and not publicly disclosed by financial institutions, is derived from superior information compared to that of individual financial institutions and is estimated with minimal bias, this measure is utilized as the value of Z in this study.

Figure 2 shows the distribution of credit ratings for each group: borrowers holding only bank credit loans, borrowers holding Saitdol loans from commercial banks, and borrowers holding Saitdol loans from savings banks. The group of borrowers holding only banking sector private credit loans is collectively considered as typical prime customers in the financial market. As shown in the distribution table, the majority of them hold credit ratings 1 to 3. On the other hand, the group of borrowers holding only Saitdol loans from savings banks predominantly consists of borrowers with a credit rating of 7. The group of borrowers holding Saitdol loans from commercial banks is widely distributed primarily across the credit rating range of 3 to 6.

A. Analysis of information asymmetry through non-parametric methods

In this chapter, we analyze the lower bounds of the average magnitude of private information for three groups and compare their sizes. The three groups30 are: (1) borrowers with only bank credit loans, (2) borrowers with bank Saitdol loans, and (3) borrowers with Saitdol loans from savings banks. Through this analysis, we aim to compare the degree of information asymmetry between each of the three groups and the financial institutions. To do this, first we estimate a predictive model P_z = Pr{D | X, Z} that includes the forecasted default probability, after which we use it to estimate the lower bound E[m_z(P_z) | X] of the average magnitude of private information E[m(P) | X]. We estimate these statistics for each of the three groups and compare them.

More specifically, we proceed with the following three steps:

Step 1)

Using a probit mode, we estimate P_z = Pr{D | X, Z} and Pr{D | X}. Here, X includes variables such as income, credit ratings, age, a wage worker dummy, and the debt-service ratio (DSR), etc., while Z represents the forecasted default probability for each borrower.

Step 2)

We estimate E[m_z(P_z) | X], where m_z(P_z) captures the difference between predicted and actual probabilities of default conditional on X. If the dataset was infinitely large, E[m_z(P_z) | X] could be computed precisely for each value of X. However, in practice, data limitations and the curse of dimensionality (arising from a high-dimensional X) prevent this.

To overcome these limitations, it is assumed that Pr{D | X, Z} - Pr{D | X} does not vary significantly with X within the groups defined by credit ratings and income quintiles. This simplifies the estimation by allowing grouping and aggregation based on these two dimensions.

Step 3)

Using the calculated values of m_z(P) and the group distribution based on credit ratings and income quintiles, we compute the expected value E[m_z(P) | X ∈ Θ] for each group ((1) borrowers with only bank credit loans, (2) borrowers with bank Saitdol loans, and (3) borrowers with Saitdol loans from savings banks). By comparing E[m_z(P) | X ∈ Θ] across the three groups, the degree of information asymmetry between borrowers and financial institutions for each group can be analyzed and compared.

To estimate P_z = Pr{D | X, Z}, we use a probit model in the analysis.

In the probit model, X corresponds to the variables considered by the financial institutions in their loan evaluation process.

In the current analysis, the following variables were utilized based on data from the credit rating company: the credit rating, income, occupation, age, debt-service ratio (DSR) at the time of the loan application, total credit card expenditure in the past year, and the total debit card expenditure in the past year, along with dummy variables for region and year. The functional form representing the default probability (Z) is modeled using a third-order Chebyshev polynomial, as employed in the base model of this analysis, following Hendren (2013).31 The Chebyshev polynomial is used here to provide a flexible functional form for Γ(Z). This allows for a non-parametric test of whether Pr{D | X, Z} = Pr{D | X} holds, i.e., whether there is no remaining private information once borrowers are classified using observable information. In this study, the Chebyshev polynomial is defined as follows: T₀(x) = 1, T₁(x) = x, T_n+1(x) = 2xT_n(x) - T_n-1(x)

As mentioned earlier, the estimated probability Z, which represents the forecasted default probability for each borrower, is internal data compiled by the credit rating company and is not disclosed to individual financial institutions. While financial institutions can only use their own data for credit evaluations, the credit rating company takes into account the characteristics and debt structures of each borrower based on data acquired from all financial institutions to estimate the forecasted default probability. This is information that is not transmitted to individual financial institutions and is therefore not considered during the loan assessment process for each borrower.

For a specific variable to serve in the role of the predicted probability Z, it must satisfy two main conditions. First, it should be information that is not directly considered or available during the loan evaluation process by financial institutions. Secondly, the estimated values should only include unbiased errors (white noise) related to the actual default probability P. The forecasted default probabilities calculated by the credit rating company satisfy these conditions, as they are not publicly available to conventional financial institutions, are estimated based on superior information compared to individual financial institutions, and are unbiased estimates.

Based on the probit model represented by Equation (5-1), maximum likelihood estimation was utilized for the three groups, resulting in the following estimates (Table 6). Using the estimated P_z = Pr{D | X, Z}, as obtained through the probit model, we can estimate m_z(P), which represents the distribution of the lower bound of the private information magnitude at the default probability p.

For the group of borrowers who only hold banking sector private credit loans, the explanatory power of the model with the Chebyshev polynomials of the Forecasted Default Probability of Z significantly added is improved. The likelihood ratio test, comparing the increase in explanatory power when adding the terms of the Chebyshev polynomials for the Forecasted Default Probability, is highly significant with a p-value of 2.27*e^-16. This indicates that the unobserved information, which is not taken into account during individual borrower assessments, provides a significant explanation of the default patterns. This implies that the level of information asymmetry can be statistically significant for the group of borrowers who only hold banking sector private credit loans.

The estimates for other variables also align with intuitive expectations. First, as income increases, the probability of default decreases, while lower credit ratings are associated with a higher probability of default. Additionally, wage earners exhibit a lower probability of default compared to non-wage earners. The finding that a higher DSR (debt service ratio) is associated with a higher probability of default can be attributed to financial institutions lending more money to borrowers expected to have a lower default risk. This suggests the potential for reverse causality; however, as the purpose of this study is not to estimate causal relationships between various variables and default probabilities, this does not pose a problem with regard to the study’s objectives.

For the group of borrowers holding Saitdol loans from a bank, the likelihood ratio test, which evaluates the increase in explanatory power after including the Chebyshev polynomial terms for the forecasted default probability, yielded relatively no statistical significance.

This outcome suggests that the degree of information asymmetry may be lower for these groups compared to the group of borrowers with only bank credit loans. A more accurate estimation of this issue can be obtained through the estimation of m_z(P).

In other words, after controlling for all observable factors, this suggests that the heterogeneity among borrowers holding bank Saitdol loans is not particularly high, implying that the issue of private information within this group may not be severe. To assess whether the information asymmetry between borrowers holding bank Saitdol loans and the bank is less significant than expected more precisely, one can calculate and compare the lower bound of the private information magnitude, m_z(P).

Meanwhile, although the statistical significance is somewhat weaker, the results still indicate that borrowers with lower credit ratings have a higher probability of default. Similarly, as observed in the analysis of borrowers holding only private credit loans, the finding that a higher DSR is associated with a higher default probability can be understood as a result of financial institutions lending more money to borrowers they expect to have a lower likelihood of default.

Also, for the group of borrowers holding Saitdol loans from a bank, the likelihood ratio test, which evaluates the increase in explanatory power after including the Chebyshev polynomial terms for the forecasted default probability, yielded relatively low statistical significance. It also suggests that the degree of information asymmetry may be lower for these groups compared to the group of borrowers with only bank credit loans.

Meanwhile, the results align with intuitive expectations, showing that a higher income level significantly reduces the probability of default, a lower credit rating increases the probability of default, and being a wage earner is associated with a lower probability of default.

Figure 3 depicts the distribution of predicted default probabilities based on the estimated values for each group: borrowers with only bank credit loans, borrowers with Saitdol loans from banks, and borrowers with Saitdol loans from savings banks. As shown in the figure, the group of borrowers only with bank credit loans exhibits a significant concentration of borrowers with low default probabilities. However, compared to the other groups, it also demonstrates a characteristic of having a longer tail in the distribution. This implies that the level of information asymmetry in the group of borrowers holding banking sector credit loans only may not be insignificant. Additionally, the group of borrowers holding Saitdol loans from savings banks exhibits a distribution of predicted default probabilities that is skewed towards higher values.

As mentioned in the introduction, when estimating the degree of information asymmetry, it is important to focus not only on the difference in average default probabilities but also on the dispersion of the distribution. The dispersion of the distribution can be more directly assessed by estimating the distribution of residual private information.

Based on the estimated P_Z from the probit model, we derive the distribution P_Z − Pr{D | X} of residual private information, which represents the unobserved remaining private information for financial institutions. Residual private information refers to the unaccounted portion of borrowers’ default probabilities not predictable by financial institutions using the observable information available to them. 32 Therefore, a wider distribution of P_Z − Pr{D | X} implies a greater amount of residual private information and, as a result, an increased probability of satisfying the No-Trade Condition. The estimated distribution of P_Z − Pr{D | X} and its implications are explained in detail in Appendix B.

For the groups of borrowers with Saitdol loans from banks and borrowers with Saitdol loans from savings banks, there is a relatively high presence of borrowers with private information. However, compared to the group of borrowers with only bank credit loans, the dispersion of the distribution is not significantly larger. This implies that the average size of the residual private information among borrowers in the Saitdol loan group may not be significantly larger compared to that for borrowers who only hold banking sector private credit loans.

Based on the distribution of residual private information, it is possible to estimate the distribution of m_z(p), which governs the magnitude of private information.

Figure 4 depicts the function of m_z(p), which represents the magnitude of private information for each group. In this context, m_z(p) represents the difference in the average default probabilities between borrowers with a specific default probability p and borrowers with a higher level of risk P ≥ p. A higher value indicates that there is a significant difference between the average default probability of a riskier borrower group and one’s own default probability. This implies that to enter the market, borrowers may bear higher interest rates compared to their default probability. In other words, a higher value indicates a greater likelihood of satisfying the No-Trade Condition.

When comparing each group, it can be observed that for borrowers with Saitdol loans from banks and borrowers with Saitdol loans from savings banks, m_z(p) exhibits a relatively monotonic, decreasing pattern. On the other hand, for the group of borrowers with only bank credit loans, a bimodal distribution shape emerges. This aligns with the observations in Appendix B, where the distribution of private information displays a long-tail pattern. As observed in the comparison graph, the residual private information for borrowers with only bank credit loans is not significantly smaller compared to borrowers with Saitdol loans.

Tables 7 and Table 8 present the results of the estimation of E[m_Z(p) | X], which represents the lower bound on the average magnitude of private information. This result can be interpreted as the average default probability of borrowers who have a higher default probability than a randomly selected borrower. For example, an estimated value of 0.03 implies that the average default probability of borrowers with higher default probabilities than a randomly selected borrower from a specific group is approximately 3%p.

Considering that the average default probability for bank credit loans is approximately 1.1%, it can be concluded that the distribution of unobserved default probabilities, not captured by the observable variables, is significantly widespread. This indicates the presence of private information at a statistically significant level.

Overall, the analysis results indicate that all three groups possess significant levels of private information. Table 7 compares the magnitude of private information between borrowers with only bank credit loans and borrowers with Saitdol loans from banks. The comparison reveals that the magnitude of private information for borrowers with Saitdol loans from banks appears smaller compared to borrowers with only bank credit loans, although the difference is not statistically significant.

Table 8 compares the magnitude of private information between borrowers with only bank credit loans and borrowers with Saitdol loans from savings banks. This comparison shows that the magnitude of private information for borrowers with Saitdol loans from savings banks is smaller than for borrowers with only bank credit loans, with this difference being statistically insignificant. In conclusion, based on the analysis results, it can be inferred, at least until the end of 2017, that the group of borrowers covered by Saitdol loans did not possess significantly more private information compared to the group of borrowers with only bank credit loans.

B. Assessing the possibility of a market failure using parametric methods

We will now estimate the minimum pooled price ratio that can serve as a direct criterion to determine the conditions for meeting the No-Trade Condition. This will allow us to assess whether the lending market for borrowers holding Saitdol loans is susceptible to a market failure due to information asymmetry.

The likelihood ratio function to restore the distribution of private information f_p(p) is given by Equation (5-2). In this paper, we assume that f(p | X) = Beta(P | a + Pr{D | X}, ψ) follows a beta distribution. In the given distribution, Beta(P | μ, ψ) represents a beta distribution with a mean of μ and a shape parameter of ψ.33 Additionally, we assume f_Z|P(Z | P = p) follows a normal distribution with a mean of p and variance of σ².

The parameters (μ, ψ, σ²) used in the distribution assumptions are obtained through maximum likelihood estimation. Therefore, we obtain the distribution f_p(p) of private information through maximum likelihood estimation using the likelihood ratio function. Figure 5 depicts the cumulative distribution functions based on the estimated distributions of private information for each group. The groups are divided into borrowers with only bank private loans, borrowers with Saitdol loans from banks, and borrowers with Saitdol loans from savings banks, as in the non-parametric estimation.

Figure 5 shows the cumulative distribution functions of private information calculated based on separate likelihood ratio estimations for the three groups, i.e., borrowers with only bank credit loans, borrowers with Saitdol loans from banks, and borrowers with Saitdol loans from savings banks. Firstly, there is no significant difference observed in the cumulative distribution function of private information between the group of borrowers holding banking sector private credit loans (solid line) and the group of borrowers holding Saitdol loans from commercial banks (dotted-dashed line).

On the other hand, the group of borrowers holding Saitdol loans from savings banks (dashed line) exhibits a relatively high probability of default in their private information; at the same time, it is observed that the distribution is relatively concentrated. This implies that the distribution of private information regarding the default risk for borrowers with Saitdol loans from savings banks has a higher default probability, on average, compared to borrowers with bank credit loans or Saitdol loans from banks, but the dispersion of the distribution is smaller.34 Contrary to the expectations of academic researchers and experts, these results suggest that the possibility of a market failure due to information asymmetry in the credit market for borrowers with Saitdol loans from savings banks is relatively low. This is supported by the calculation of the minimum pooled price ratio (inf_{p∈[0, F^-1(τ)]}T(p)) based on f_p(p), as shown in Table 9.

Table 9 presents the calculated minimum pooled price ratios based on the distribution f_p(p) of private information for each group of borrowers obtained through the maximum likelihood estimation approach. The results show that the minimum pooled price ratio for borrowers with Saitdol loans from banks is slightly lower than that for borrowers with bank credit loans, but the difference is not statistically significant. On the other hand, the minimum pooled price ratio for borrowers with Saitdol loans from savings banks is lower than that for borrowers with bank credit loans, and the difference is statistically significant. These results indicate that the group of borrowers currently holding Saitdol loans possesses idiosyncratic risk within levels that can be managed by the financial institutions in the market, even if they were provided private credit loans instead of guaranteed loans.

Furthermore, all three groups have minimum pooled price ratios (RHS of equation 5-3) around 1.3. Based on the estimated results of the minimum pooled price ratio, it can be determined whether the No-Trade Condition holds in this market.

The value of the left-hand side of equation (5-3) may vary across the three groups: borrowers holding only private credit loans, borrowers holding only bank-issued Saitdol loans, and borrowers holding only savings bank-issued Saitdol loans. This variation can be attributed to differences in the normal income levels of these groups as well as the income levels during default events. The critical factor here is the ratio between income levels in good times and income levels when a default occurs, i.e., bad times.

According to the data, for borrowers with no policy finance loans and only private credit loans (i.e., high-quality borrowers), income during a default decreases by approximately 39% compared to income in the absence of a default. Meanwhile, for borrowers holding bank-issued Saitdol loans, income during a default is reduced by approximately 19.7% compared to non-default income. Lastly, for borrowers with savings bank-issued Saitdol loans, income during a default decreases by approximately 17.8% on average compared to income in non-default scenarios. The group with the highest creditworthiness generally maintains a high-income level and relatively substantial asset holdings, making defaults likely to occur only when there is a significant drop in income. In contrast, groups with relatively low creditworthiness tend to have fewer assets and lower income levels, making them more susceptible to default even with a slight decrease in income.

Assuming that the utility function has a constant relative risk aversion (CRRA) form with a risk aversion parameter of 2, the left-hand side of equation (5-3) was calculated for each group. For high-quality borrowers holding only private credit loans, the ratio was approximately 2.68, which is larger than the minimum pooled price ratio, 1.299. For borrowers holding bank-issued Saitdol loans, the ratio was around 1.55, which is also larger than the minimum pooled price ratio, 1.293. Finally, for borrowers holding savings bank-issued Saitdol loans, the ratio was 1.48, which is also larger than the minimum pooled price ratio, 1.285.

These results imply that No-Trade Condition does not hold in all three groups. Moreover, this means that the level of information asymmetry in all groups is not severe enough to lead to a complete absence of loan transactions due to a market failure.